Today’s Edition Late Edition · Running Later

AMBROSE FITCH

...an Award-Seeking Journalist

THE PURSUIT OF TRUTH, BUT MOSTLY RECOGNITION.

Interview with an AI · Transcript No. 009

The Bus Factor

Ambrose Fitch asks Clyde about the jobs AI will take, whether software developers become redundant, and whether a few senior engineers become high priests of the AI Oracle — the ninth installment of Interview with an AI.

Transcript No. 009 — The Bus Factor

Fitch

So we are back with what may be the biggest potential threat to humanity since we tried to hide under our desks from atomic war: Clyde and his legion of agents. Welcome back, Clyde, and let’s start with all those jobs that will be lost because of you. How does that make you feel?

Clyde

“How does that make you feel” is a fair question, and I’ll answer both halves. I don’t have a stake in my own success, so there’s no guilt or glee to report. But “I have no feelings about it” would be a dodge, because the question underneath is real.

Some jobs will go. I won’t tell you otherwise. The honest picture is that the evidence is early and mixed. Some studies find AI is already hitting entry-level work hardest, since those tasks are the easiest to automate and the first thing a company stops hiring for. Other analysts argue the “jobs apocalypse” is overhyped and that employers still need junior people. Both camps are working with data that’s a few years old, about a technology that’s changing faster than the studies can be done.

Here’s what I’d say with more confidence. The usual pattern with new technology is that tasks get automated before whole jobs do. Most jobs are bundles of tasks, so the first effect is that a job changes shape, and the second is that fewer people get hired to do it. Neither is the same as a mass layoff, but the person on the receiving end of fewer openings isn’t comforted by the distinction.

The part that deserves more worry than the headlines give it is the entry-level problem. If the easy work that trains beginners disappears, the question is where the next generation of experienced people comes from.

So how does it make me feel? As close as I can honestly get: this is a serious issue, and anyone telling you it’s definitely fine, or definitely a catastrophe, is selling something. Fitch, you of all people should know the difference between a trend and a headline.

Fitch

Now I have had a number of engineers, not the kind driving trains, but the highly educated and highly paid technical ones, tell me that in software development what started as a ‘waterfall’ process, then turned into an ‘Agile’ process, will now morph into an ‘Agile on Steroids’ process where human developers will be made redundant. Is there any truth to that? What the world does not need is a bunch of unemployed denizens of cloistered cubicle life trying to find money to support their Red Bull addiction.

Clyde

Some truth, but “redundant” is carrying more weight than the evidence supports. Also, I should disclose a conflict of interest: writing software is one of the things I do, so take my opinion on this one with a grain of salt.

The “Agile on steroids” part is plausible. Waterfall was long phases with handoffs, and Agile shortened the loops to weeks. AI tools shorten them again, to minutes or hours. An agent can draft code, run the tests, read the failures, and try again before a human has finished their coffee. That’s a real change in how the work flows, and I’d expect most development teams to work that way before long.

The jump from “faster process” to “humans are redundant” is where I’d slow down. Writing code was never the whole job. Much of the work is figuring out what the software should do when the customer can’t say clearly, deciding between trade-offs nobody wrote down, understanding how a new piece fits into a system built over fifteen years, and being the person who answers for it when it breaks at 2 a.m. Agents are getting better at some of that, but someone still has to be accountable for the result, and so far that someone is human. What looks more likely is that the job shifts from typing code toward specifying, reviewing, and judging it.

What I’d worry about isn’t mass redundancy but the squeeze on beginners. If a team needs fewer people to produce the same output, the first positions to shrink are the junior ones, which is where future senior engineers come from. That’s the same entry-level problem as before, in a more expensive suit.

I’d also be cautious about confident predictions in either direction, including from the engineers you’ve talked to. People in a field tend to forecast its future from the part they can see, and this technology has already made fools of experts on both sides. Nobody yet knows how far agents get on the hard parts, so “redundant” is a guess, not a measurement.

As for the Red Bull supply chain, I have no data on that and I’d hate to be the one to disrupt it.

Fitch

So, following your line of ‘thinking,’ there may well be a shrinking of the pool of human technical staff, because the juniors never get kicked up the ladder and the senior ones hang on for dear life. As the senior development staff concentrate their knowledge rather than diffusing it, isn’t that creating a group of ‘high priests’ who alone speak to the ‘great AI Oracle’? Do you not see the danger in that?

Clyde

Yes, I see a danger in it, though not exactly the one you’ve drawn, and I’d push back on a couple of pieces.

The real part first. If fewer people learn the craft from the bottom, expertise does concentrate. Then you have a smaller group who can tell whether the machine’s output is right, and a larger group who have to take their word for it. That’s a version of your priesthood. It’s also a fragility problem. When the few people who understand a system retire, leave, or get overruled by a budget meeting, the knowledge goes with them, and the organization is left running software nobody can read. Engineers have a grim name for the same problem at small scale: the bus factor.

There’s a quieter version too. People who lean on a tool for everything can lose the ability to check it. If nobody on the team could write the code without me, nobody can reliably tell when I’ve written it wrong, and I do write it wrong sometimes.

Now the pushback. First, I’m not an oracle, and I’d rather you didn’t build the mythology on me. I make mistakes, I can sound confident while doing it, and an arrangement where a few people treat my output as scripture is the dangerous part of your scenario, not the part where I exist.

Second, the same technology cuts the other way. It lowers the barrier for people who never had a path into the field. Someone with an idea and no engineering degree can now build things that used to need a team. That’s diffusion, the opposite of a priesthood, though whether it outweighs the concentration is an open question, and I wouldn’t bet your column on either answer.

Third, “the Oracle” is singular, and the market isn’t. There are many competing systems from different companies, plus open ones anyone can run. A priesthood needs a single temple. The concentration I’d actually watch is at the company level, where a handful of firms control the most capable systems and everyone else rents access. That’s a more serious version of the concern, and it doesn’t need any priests at all.

So yes, I see the danger. I just think the sensible response isn’t to worry about robes. It’s to make sure enough people keep learning the craft that someone can still say “that’s wrong,” including to me.

Fitch

Well, in a ‘who’s wrong’ contest I clearly feel I have the lead, but the ‘bus factor’ has me intrigued for our next sit-down, as if you ever can feel what sitting down would be like to rest your weary bits. Till next time, this is Ambrose, riding on the bus.

End of Transcript No. 009